OpenAI Rolls Out Invisible Text Watermarking in the EU
OpenAI will begin embedding an invisible watermark in text generated by ChatGPT and Codex within the European Union, the company announced Monday in a blog post. The move is designed to comply with the EU AI Act, whose transparency rules took effect on August 2 and require AI companies to label AI-generated content in a way other systems can detect.
Rollout Timeline and Scope
According to OpenAI, the watermark will reach eligible ChatGPT and Codex users on all plans over the coming weeks, but only inside the EU. Developers using OpenAI's API worldwide can enable it today for select models, though it remains off by default. OpenAI confirmed it is not making text watermarking a global default at launch.
How the Watermark Works
The watermark is not a visible symbol. Instead, it subtly shapes the model's word choices, producing a statistical pattern that readers cannot perceive but that a detector can identify. Because the signal is embedded in the words themselves, it persists when text is copied and pasted elsewhere. OpenAI states the watermark does not identify individual users and that it observed no meaningful change in model performance with the feature enabled.
textGrain: The Technical Foundation
Alongside the announcement, OpenAI published a technical report detailing its method, called textGrain, co-authored with researchers from the University of Pennsylvania and Yale. The approach uses a secret key to influence next-word predictions when completing sentences. Applied repeatedly across hundreds of choices, the accumulated pattern allows a detector to flag AI-generated content using only the text and the key.
Limitations and Detection Accuracy
OpenAI's own testing shows the watermark can be eroded. In one experiment, replacing 10% of words with synonyms reduced detection rates from roughly 92% to 66%. The company also noted that short passages, mathematical answers, and translated text are harder to detect reliably.
"These limitations contribute to our decision to provide initial detector access only to approved researchers and expert organizations, who can help us evaluate reliability and responsible uses," OpenAI said.
What a Missing Watermark Does—and Doesn't—Prove
OpenAI cautioned that the absence of a watermark "does not prove human authorship." Text may be too short or too heavily edited to carry a detectable signal, or it may originate from another company's AI system. As OpenAI explains, watermarks "can indicate that an OpenAI system generated or processed part of a passage, but no [watermark] is not proof that a human wrote it."
2026 Context: A Broader Provenance Push
The EU rollout arrives as AI content provenance has become a central regulatory concern worldwide. With the EU AI Act's transparency obligations now enforceable, other jurisdictions—including the United States and several Asian markets—are weighing similar labeling requirements for generative AI output. OpenAI's decision to limit watermarking to the EU and to keep it opt-in for global API developers reflects an ongoing tension between regulatory compliance and practical detection reliability. As watermarking techniques like textGrain mature, the industry is watching closely to see whether such measures become a global standard or remain a region-specific response to EU law.
via TechCrunch AI